Looking for HV fets for direct drive

[maudio]

yes that would be ideal. But unfortunately an esl is not behaving like a resistor, but like an almost perfect capacitor.

Placing resistors in parallel might make things a little easier when it comes to stability of the feedback loop of your amp, I see no other advantage. And the price is that the amplifier has to deliver even more power which is only wasted to heat in your resistors. Things are hard enough as it is. And for unstable amplifiers there are much better solutions.

 

[bentoronto]

I'm sure that is good advice for somebody. But in my universe (which includes actually building and running the amp for two decades), I'll take a nicely behaved resistor load any day.

Ben

 

[jcx]

Class A direct drive for large ESL isn't impossible - just daunting - may need to run a dedicated 240 V line for the power

push-pull can be done with a "SRPP" derived modulated CCS

Class AB does require hi V drive for the upper follower device if quasi-comp n-channel output is used since they are the higher V rated parts - xfmr gate drive does seem like an option but could limit loop bandwidth

 

[maudio]

Class A direct drive for large ESL isn't impossible - just daunting - may need to run a dedicated 240 V line for the power

everything can be done, it's just not practical... Imagine getting rid of 3500 W of heat in a safe way. You need a lot of components to distribute that heat. It is almost impossible to achieve sufficient isolation between power fets and heat sink, so all fets need their own separate heat sink and all heat sinks have to be hidden within the enclosure as they will not be safe to touch. You need a huge enclosure. Forced ventilation must be used and a lot of it, causing noise disturbing your superb music experience. Your electricity bill will skyrocket, maybe not such an issue in the USA but it is here in Europe where energy prices are several times higher.

push-pull can be done with a "SRPP" derived modulated CCS

Will in an ideal situation reduce dissipation by a factor 2, still a lot of heat. Ideal situation means purely resistive load and we don't have that. It is not easy to get SRPP to work well over sufficient bandwidth with a reactive load and even if you do get it to work the reduction in dissipation will be a lot less than a factor 2 due to phase shift issues.

Class AB does require hi V drive for the upper follower device if quasi-comp n-channel output is used since they are the higher V rated parts - xfmr gate drive does seem like an option but could limit loop bandwidth

Bandwidth issues seem solvable to me. But I see a much bigger problem in the parasitic capacitances between prim/sec windings of the transformer. The full output swing is present over this capacity. And if the voltage is high enough even a few pf will cause enough current flow to mess things up. As a result you to loose the ability to drive the upper follower accurately enough. Don't ask me how i know
Same applies for the use of optocouplers btw.

 

[invisible force]

Hello maudio,
do you mind revealing some thoughts on how to include a step up in the amplifier feedback loop ? Mr. Wright (Dayton Wright,look at XG 10 Mk III error_correcting-ESL) followed this solution stating incredible low THD numbers...
Concerning parasitic capacicty problem with optocouplers for galvanic insulation of gate signal in a dd amp, you are - as usual - right up the point : take for example 1pF with 1kV of AC Voltage drop - current is the same as 1nF with 1V AC drop...
regards, Philipp

 

[jcx]

common plastic DIP optoisolators may not survive multi kV for long - you need specialty products

transformer gate drive isolation amp with DC to low MHz bandwidth probably requires modulation/demodulation with >10 MHz FM, or PWM thru the xfmr

at that circuit complexity level you might as well add non-switching geometric mean Class AB circuitry too

few pF pri-sec is hardly noticible in a drive current budget that includes nF panel C

 

[bentoronto]

Hello maudio,
do you mind revealing some thoughts on how to include a step up in the amplifier feedback loop ? Mr. Wright (Dayton Wright,look at XG 10 Mk III error_correcting-ESL) followed this solution stating incredible low THD numbers... Philipp

I'm not sure what I said or didn't say before. Here's what I do say:

Starting with a biggish drive signal followed by a mild 1:10 step-up transformer (instead of the usual 1:75) might be a mighty good way of reducing transformer distortion and easily feasible. Sometimes a hybrid solution is a pretty good solution, even if not elegant.

I don't know if Wright had any feedback from the secondary side but it might be more feasible than with the challenges of getting a feedback-worthy signal from the far side of a 1:75 transformer.

My direct-drive amp (which drove a bunch of resistors and, by the way, a little parallel capacitor AKA the ESL load) did have total feedback.

I seem to recall that a bit of amp distortion sounds better than a bit of transformer distortion, given a choice.

Ben

 

[maudio]

Hello maudio,
do you mind revealing some thoughts on how to include a step up in the amplifier feedback loop ?

we are getting a bit offtopic, excuses to the TS. I'll try to keep it short...

The idea behind the transformer in a feedback loop developed when I had switched from building DD amps (fun but pointless) to winding my own transformers. What I learned from that is that an ESL transformer is always a terrible compromise. You want high stepup to get decent output but increasing stepup decreases bandwidth and results in lower impedances at the high side of the audio range. With a stepup of 1:100 or higher impedances of 1 ohm or less at 20khz are not uncommon. That has a rather negative impact on sound quality, amplifier distortion increases steeply with increasing output current (beta droop in the output devices). Especially at higher frequencies where openloop gain is already lower so there is less feedback to compensate. This is imho the main reason why esl’s with low stepup ratios often sound so much cleaner.

What you ideally want is :
1. reduce to load on the amplifier, that should ideally not drop below 6 ohm or so.
2. do something about the non-linearities in the transformer (caused by core physics and other boring stuff)
3. increase stepup if possible

The solution I found is to place a LR network before the transformer. This radically transforms the frequency response and impedance of the transformer, see picture. Top curve is a normal transformer + esl, bottom is the same transformer with the same esl but now with 10 ohm//2.5mh in series with the primary. The frequency response now starts to drop at approx 2-3 khz with a nice 6db/oct slope. Transition to 12 dB does not occur before well above 200 khz. Even better, the impedance seen by the amp does not drop anymore at higher frequencies.
The price is the amp has to deliver more voltage at high frequencies to compensate for the dropping response. Basically we are trading current demands for voltage demands but that is a good deal. The energy content of music drops steeply at high frequencies, so there is plenty of headroom in the voltage department.

Now we place a feedback loop around the whole thing. If we do that correctly and we add enough loop gain we get a nice straight frequency response to above 20 khz, straightening out any transformer non-linearities along the way. Those are most pronounced at low frequencies and wow, there we have plenty of feedback. Two beers for the price of one

The challenge is to get the feedback loop stable. You have to use nested feedback loops, using the frequency response of the transformer/RC network to set the dominant pole in the outer feedback loop. It requires a power amp with a bandwidth of at least 1 MHz, to keep enough phase margin within the loop. All of this is no big issue with modern components, in fact with a good design there is even room to increase stepup to 150-200.

I built a good sounding prototype with a self-made 1:125 transformer that reaches 40khz bandwidth driving 500 pf and I believe that is quite impressive.
The whole project got a bit dusty lately as life puts other demands on a man but I plan to pick it up again using 6/230V toroids, got tired of winding. If anything good comes out I'll post it here.

So much for my attempts of keeping it short

 

[invisible force]

This indeed looks like best alternative to dd drive. The high voltage dividers will need bypass caps. A circuit diagramm of the two long tailed pairs and compensated HV dividers would be great...
One problem is haunting me for a long time : the two primary - secondary capacities of the step up will leed to considerable current into the main earth..further deatails can be extracted in the very worthwhile reading white paper AN004 (<hum & buzz...> JENSEN TRANSFORMERS, INC. - APPLICATION PAPERS AND SCHEMATICS) by Bill Whitlock. This should be valid for all ESL´s with step up connected to main earth...
great idea maudio!

 

[invisible force]

A suitable high voltage N channel FET for dd drive : STW3N150 (STP3N150 - STMicroelectronics) by STMicroelectronics..
A comment to my last post : The problem with the current into the main earth from the step up center tap is, that it will lead to a voltage drop across the cable shield resistance of preamp - power amp connection (also cd - preamp etc.) - this is not 50Hz or so, but phase shifted fullrange music signal spectrum, which is seen simply as input signal by the receiving amp..signal gets covered by "fog"..distortion..

 

[invisible force]

just comes to my mind : as it is the difference of the two step up primary-secondary caps which is causing the "fog current" it should be possible to reduce it by adding a cap to the smaller one...it needs to be a high voltage part and it will not be a perfect match because of the boring transformer complexity...sorry if I am a bit off topic, but it should be valid for most ESLs...
regards, Philipp

 

[maudio]

Back on topic

A suitable high voltage N channel FET for dd drive : STW3N150 (STP3N150 - STMicroelectronics) by STMicroelectronics..

Ciss almost 1000pf, that is a bit high. One of the most important parameters to look for is low gate capacitance (Ciss). You have to stack multiple mosfets to handle the voltage and to do so you need a resistor divider driving the gates. Those resistors together with Ciss limit your bandwidth. Lowering the value of those resistors makes your dissipation skyrocket.

The 1000V IR IRFBG20 is such a nice mosfet for this application because it has only 500 pf Ciss.
The BFC40 from Semelab has 550 pf and will even handle 1500V. Semelab also has the BFC60 with as little as 40 pf but that will only handle 100mA/20W and that is just not enough for a 4kV amp. Ideally would be a mosfet in between those 2 but I have never found one. Come on guys, start producing a bfc 50

The higher the current rating the higher Ciss will be. So you need to get the smallest available mosfet that will handle the power.

 

[invisible force]

O.K. STP2NK100Z (STP2NK100Z - STMicroelectronics) is a better choice (Ciss 500pF / Vds 1000V) and it is protected by build in zeners..
Maudio : I regard your feedback solution far more practicable than dd drive...would be great, if you find the time for a circuit diagram of electronics surrounding the power amp...

 

[invisible force]

If you look at Patent DE102008034456 (Espacenet - Original document)
you see a solution of dd drive with optocouplers..the inventor Fritz Linck is a very friendly man - I visited him to listen to his dd driven Quads...well it was a working solution but imho sound was worse than done with transformers..sorry to say that..For a short time it was offered as an Quad upgrade option (€ 10.000) by the official Quad distributer in Germany..but they removed it for sound reasons.

 

[invisible force]

No comments maudio..??

 

[jcx]

cascading AC performance can be improved with added C divider in parallel with the R

Cgs is bootstrapped by the device gain and appears reduced in value, Crss becomes an issue depending on the MOSFET type but the AC divider C may be smaller than you expect and don't add too much load

but even tweaked for better AC performance you can have catastrophic failure once one device goes, lower number of series devices is safer - KOSS E/90 electrostatic amp uses 3x 500 Vds MOSFET with +/-600 Vsupply and zener clamps them D-S

IXYS advertises up to 4 kV MOSFET

 

[alexberg]

cascading AC performance can be improved with added C divider in parallel with the R

Cgs is bootstrapped by the device gain and appears reduced in value, Crss becomes an issue depending on the MOSFET type but the AC divider C may be smaller than you expect and don't add too much load

but even tweaked for better AC performance you can have catastrophic failure once one device goes, lower number of series devices is safer - KOSS E/90 electrostatic amp uses 3x 500 Vds MOSFET with +/-600 Vsupply and zener clamps them D-S

IXYS advertises up to 4 kV MOSFET

Every modern Si MOSFET has build-in quite powerful zener with energy up to few Joules, somewhat like avalanche diodes. Good luck with external clamp, unless you need to do so below Vds max. Moreover Zeners have smaller dies... Good HV MOSFET can sustain single overload around 20kW @ 25us and even more @ lesser time.
So no chain breakdown, just keep supply bypass capacitor value reasonable. And yes, biggest problem is Crss which feeds back to gate thus demanding rather low R values in ladder divider and/or capacitive shunts across the resistors. Such measures negatively affect achievable gain and HF gain in particular.
Have a look at
http://www.pes.ee.ethz.ch/uploads/tx_ethpublications/Sic_Switch_Upload_01.pdf
Such circuit is even more demanding in regard to dynamic performance in respect to audio amp.

 

[invisible force]

gate voltage divider done with avalanching diodes is for switching purpose only...no linear application possible by doing this...interesting, but not for audio..have fun..

 

[maudio]

If you look at Patent DE102008034456 (Espacenet - Original document)
you see a solution of dd drive with optocouplers..

It is quite an achievement to get something like that to work at all, to make it performe good is almost impossible. There are several issues with optocouplers. First the are very noisy making it difficult to get sufficient s/n ratios. Second and worse they are not very linear. The linearity can be improved a lot by using linear optocouplers that use a second photodiode to compensate the non-linearity’s with a feedback loop controlling the led current, something I do not see in this design. But even then it is difficult to make a good sounding amplifier with optocouplers in the signal path. And there is the safety/reliability problem but that might be solvable choosing the right components.
I once built an experimental optically coupled output stage running from 2 kV with irfbg20mosfets, using hcnr 200 linear optocouplers. It did not work very well, in fact it proved almost impossible to get it to work at all. The output stage itself could oscillate with the inputs of both optical circuits shortened, no feedback whatsoever. Even when I managed to get it stable (at the cost of much needed bandwidth of course) I still had very poor control over the top fet, resulting in open loop performance of the output stage that was so horrible non-linear that I abandoned the idea altogether. No chance of building a good sounding amplifier from such a basis.
The reason is simple once you know it: we have a photodiode in the optocoupler that is generating photo currents in the nano and microampere range, coupled to a highly sensitive circuit that converts these minute currents to a control voltage for the mosfet . Now we have approx 0.5 pf capacitance between input and output side of the optocoupler (that no spice model takes into account btw). For the top optocoupler we have full output voltage swing over that capacitance, injecting error currents. Do the math yourself... Now add to the equation additional capacitances introduced by the DC/DC converters needed to power the top circuit and you don’t stand a chance to make it behave in any decent way.

cascading AC performance can be improved with added C divider in parallel with the R

Cgs is bootstrapped by the device gain and appears reduced in value, Crss becomes an issue depending on the MOSFET type but the AC divider C may be smaller than you expect and don't add too much load

I experimented a lot with output stages with mosfets and found little room for improvement using compensated HV dividers. The internal capacities of the mosfets are not only rather large but behave highly "dynamic", they change radically with voltage changes. You can safely forget any values from the datasheet. In order to get any improvement from a compensated divider you need to outnumber the internal capacitances requiring very large capacitors, couple of nf at least. As all this extra capacity is basically parallel to the output device it adds so much load that things just get worse overall.

but even tweaked for better AC performance you can have catastrophic failure once one device goes, lower number of series devices is safer

Not only safer but also faster. Reliability can be improved with fast diodes between gate and source, to prevent breakdown of the channel during negative transients. Zeners are not recommended as they are slower and not needed, positive transients are no problem as the fets will protect themselves in that case. Using simple 1n4148 I did not ever encounter reliability problems on output stages using 4 stacked irfbg20 mosfets for each device, even when processing square waves and other nasty waveforms.
Stacking more fets is not only ruining reliability but also bandwidth, even with 4 fets openloop bandwidth of the output stage was already far below 20 khz. Did some test with large number small 300V fets but found openloop bandwidths of 1-2 khz, no good starting point.

 

[invisible force]

Yes, linear optocouplers have been used (could be the hcnr 200..), the floating operational amps are types from AD (grabbing from my brain), specially designed for driving capacitive loads (remaining stable by enabling the internal compensation cap becoming bigger as capacitive load is heavy..)...the inventor told me he spent 10++ years on the project to get it work...I believed him...
Your feedback solution is more practicable, maudio...